Is My Car Evil? A Review of Non-Anthropocentric Theories of ......Is My Car Evil? A Review of Non-Anthropocentric Theories of Moral Agency 107 Introduction January 22, 2018. It is

107Is My Car Evil? A Review of Non-Anthropocentric Theories of Moral Agency

Introduction

January 22, 2018. It is Monday morning and a

huge firetruck is stationing in the middle of the

Interstate 405, in California. A car, travelling at

100 km/h, proceeds right toward the truck

without slowing down. Nothing obstructs the

view of the driver. There is no malfunctioning

in the electrical system and the hydraulic

brakes are ful ly operative, ready to be

activated. Yet, the driver does not push the

brake pedal. The car continues its run, getting

closer and closer to the firetruck parked in the

centre of the road. The impact seems now

inevitable. The vehicle crashes into the back of

the red truck. The metallic front of the car

folds in on itself. Surprisingly, the driver is safe.

This time nobody got injured.1

More than 1 million people die every year in

traffic accidents all over the world (WHO 2015).

However, quite curiously, this banal collision

caught the attention of some of the most

popular newspapers around the globe. Why

was this relatively harmless incident taken so

seriously by the press? One of the reasons is

that, apparently, the man sitting at the driver

seat was not actually driving the car during the

accident. But not because he fell asleep at the

wheel, nor because he was distracted by a

no t i f i ca t i on f rom h i s smar tphone . He

deliberately chose not to drive. Indeed, someone

else was driving the car in his place. But here

is the issue: that someone, the entity that was

really behind the wheel during the accident,

was not a human being. It was the “autopilot”.

It was a piece of software.

Nowadays smart technologies substituting us

and acting for us are everywhere. We are

quickly getting used to delegating everyday

tasks to a multitude of artefacts. Artefacts that

constantly mediate our experience of reality

and help us in the process of making decisions.

Yet, from a moral point of view, we do not

really know how to consider all these entities.2

What can be said about the above-mentioned

software that destroyed a car and that could

* University of Tokyo, Graduate School of Interdisciplinary Information Studies, ITASIA D2

キーワード：Actor-network theory, Information ethics, posthumanism, moral agency, responsibility, morality of technology.

Is My Car Evil? A Review of Non-Anthropocentric Theories of Moral Agency

Tommaso Barbetta*

108 東京大学大学院情報学環紀要　情報学研究　№96

have killed its owner? Is it responsible for the

accident? Is the company that designed it

responsible? Or should we rather blame the

human driver, who was supposed to - but

presumably did not - keep his hands on the

steering wheel and be vigilant in order to take

control of the car before the collision. And what

about the algorithm that in 2012 made Knight

Capital lose 440 million USD in 45 minutes?3

Can it be blamed for the disruption of the

company? Nobody designs trading algorithms

with the intention of breaking a company apart

or causing a market to crash, but that is

nonetheless what they ended up doing in some

cases (MacKenzie 2014, 3). On the surface,

(quasi-)autonomous technologies appear to

behave as if they could make decisions and act

on their own. However, the lack of intentions -

besides those delegated by humans - behind

such Boolean logic, has led consequentialists to

cast serious doubts over the significance of

recognizing the actions performed by artefacts

as morally charged (Peterson and Spahn 2011).

Due t o the i r devo t i on t o the i dea o f

intentionality, standard ethical theories have

been unable to recognize the moral implications

of information and communication technologies

(ICT) such as the one adopted by the above-

mentioned autopilot.

Outside the field of ethics, over the last 30

years several non-anthropocentric theories of

agency have emerged. In particular, actor-

network theory (ANT) has provided an

effective framework for the assessment of the

role played by artefacts, and other kinds of non-

human entities, in our society.4 However, it is

not yet clear if and how a notion of agency

detached from intentionality, such as that of

ANT, could be translated to ethics. Once it is

freed from intentionality, is moral agency a

characteristic of a specific category of entities,

such as humans, living beings or algorithms, or

could it be ascribed to anything? Furthermore,

how could the question of responsibility be

reframed in order to fit an ontology according

to which agency is always distributed among

multiple actors?

After a brief overview of the instrumentalist

approach to technology embraced by standard

anthropocentric moral theories, the present

paper investigates these questions by reviewing

the l i terature o f ANT (a long with i t s

postphenomenological adaptation). Furthermore,

in the final sections, the article turns to

information ethics (IE), a non-anthropocentric

ethical theory influenced by computer science,

whose different definition of moral agency

might help us to solve some of the difficulties of

an ANT-informed ethics.

1 Instrumentalism


From the po int o f v iew of s tandard

contemporary ethical theories , such as

d e o n t o l o g y , c o n t r a c t u a l i s m a n d

consequentialism, it would be incorrect to

consider an artefact as a moral agent. In

standard ethics by definition an action is moral

only as long as it is initiated by some kind of

intentional state of mind, i.e. only if it is the

product of a conscious reason (Himma 2009,

Koops et al. 2010). Given this definition, there is

no way my car could be a moral agent since,

due to the limits of our current technologies, a

computer cannot possibly have intentional

states of mind.

By accepting intentionality as a necessary

condition for moral action, standard ethical

theories therefore set a clear separation

between human entities, which reflect, make

decisions and have intentions and free will, on

the one hand, and purely neutral artefacts, on

the other. To rephrase the (in)famous slogan of

the American National Rifle Association: cars

don’t kill people, people kill people. An artefact

– be it a car or a gun - is only a mere

instrument, and as such it has always different

possible uses. It is ultimately the human user

who determines what to do with it (Pitt 2014).

Responsibility lies always in a human.

Because of their commitment to the idea of

moral responsibility, standard ethical theories

are incapable of recognizing many of the

pressing moral issues of our information

society. First of all, a strong anthropocentric

approach seems to ignore the existence of

artefacts that are able to learn from the

environment, correct themselves and make

decisions that were not intended either by their

producers, or by their users. Take the case of

the autopilot mentioned in the introduction,

which relies on a computer vision system: this

system was built on an artificial neural network

where code is not entered line by line by

human programmers and whose output is often

obscure to the original creators themselves. It

wou ld be inaccurate to ho ld a human

programmer responsible for the individual

choices made by such kinds of software

(Matthias 2004).

Moreover, standard ethical theories seem to

downplay the role of technologies in mediating

the state of mind of individual humans. The

increasing adoption of neuromarket ing

strategies attests the widespread awareness

among commercial corporations of the influence

that ad-hoc physiological stimuli might have on

the behaviour of consumers. For example, the

layout and atmospherics of most casinos are

designed to make customers move following

specific patterns and keep them playing as long

as possible (Schüll 2012). This kind of artificial

environment is by no means neutral. Rather, it

actively manipulates the perceptions of its

users. What we call addiction is not the

unidirectional product of the harmful decisions

made by an (genetically) impaired brain. It is a

relational phenomenon involving a human and


an environment often - but not necessarily -

designed to make use of his limitations.

The following sections investigate advantages

and disadvantages of the adoption of a notion of

agency influenced by ANT, focusing on the

moral issues raised by the crash.

2 Actor-network theory

Developed in the early 1980s in the field of

sociology of scientific knowledge, ANT is a

heterogeneous set of linguistic categories that

initially emerged to trace the active role of non-

human entities in the analysis of scientific

practices (Latour 1983). The common ground,

shared by the social scientists adopting this

analytical framework, is a non-anthropocentric

view of reality based on the concept of

network. According to Bruno Latour, the

author who more than anyone else has unfolded

the philosophical consequences of ANT, the

best way to take into consideration how a

heterogeneous set of non-human entities, from

microbes to airbags, participates in our

collective life is to adopt a “network-like

ontology” (Latour 1996, 370).

The next sections investigate the ethical

implications of this ontological stance. Following

a review of the key concepts of symmetry and

mediation, the article discusses how the notions

of “prescription” and “distributed agency”

might inform a moral analysis of non-human

entities.

2.1 Principle of symmetry

ANT is based on the idea that there is an

ontological symmetry between humans and

non-human entities. This is sometimes referred

to as the “principle of generalized symmetry”

(Cal lon 1986) . Human beings , biological

organisms, material objects and abstract

entities are all part of the same reality. Given

this principle, a common terminology, one

which is not b iased by the previous ly

mentioned human vs. non-human dichotomy, is

needed by social scientists. “Actor”(or “actant”)

is the term used to express the minimum unit

of our reality. An actor is literally anything that

acts, any node that contributes to a network.

An actor is defined by its action, and since

“there is no other way to define an action but

by asking what other actors are modified,

transformed, perturbed or created” (Latour

1999, 122), it follows that an actor is necessarily

defined by its relations with other actors within

a network.

According to John Law (1992), from an

analytical standpoint this symmetrical approach

leads us to a radical rejection of any difference

in kind between humans and objects. Humans

are not at the centre of the universe anymore.


They are equal to any other entity. However, it

would be a mistake to read ANT’s analytical

norms as moral norms. Indeed, justifying a

moral symmetry on the basis of an ontological

symmetry would be logically flawed: the fact

that reality is in a certain way does not imply

that we ought to act accordingly - something

which was pointed out already by Hume (1896).

“We need, I think, to distinguish between ethics

and sociology. The one may - indeed should -

inform the other, but they are not identical. To

say that there is no fundamental difference

between people and objects is an analytical

stance, not an ethical position. And to say this

does not mean that we have to treat the people

in our lives as machines.” (Law 1992, 383). Law

is – rightly – afraid of the consequences of a

misinterpretation of the principle of symmetry.

Yet, he leaves the door open for a possible

contribution of ANT to ethics.

I n a n a r t i c l e t i t l e d “Mor a l i t y a nd

Technology”, Latour (2002) takes on this task

and attempts to elaborate an ANT-informed

view on ethics. In the article, Latour openly

criticizes standard ethical approaches that

divide technology and morality into two

separated realms, that of means, on the one

hand, and that of ends, on the other. Artefacts

cannot be reduced to mere means, since they

lay the conditions for our actions, and thus

mediate our behaviour. According to Latour,

wha t we c a l l “human” c ann o t e x i s t

i ndependent ly f rom the techno log i ca l

mediations it is intertwined with (Latour 2002,

252). “Generalizing the notion of affordance, we

could say that the quasi-subjects which we all

are, become such thanks to the quasi-objects

which populate our universe” (Latour 2002, 252-

253). We, as quasi-subjects, are free either to

accept or reject the programmes of action - the

affordances - embodied by the artefacts

surrounding us, but nonetheless our behaviour

is necessarily mediated by them. A moral

approach informed by the relational ontology of

ANT, is therefore one that focuses on technical

mediations.

2.2 Mediations

In ANT the way actors interact with each

other is referred to as mediation. Two actors,

two nodes of a network, always interact with

each other by the means of a third actor

(Latour 1996, 378). Given an actor A and an

actor B, their interaction necessarily occurs

through the mediation of an actor C. This third

actor C is a “mediator” (Latour 1996, 373). In

the case of non-autonomous cars, for example,

human drivers control the wheels using the

steering wheel. At this level of abstraction, the

steering wheel is a mediator. However, the

steering wheel is merely an interface and there

is a series of other mediators between it and

the wheels: e.g. a steering column, several

sensors, an electronic control unit and a motor,


all translate the mechanical torque of the

steering wheel operated by the driver into the

actual steering. Moreover, it has to be

remarked that this is not a one-dimensional

chain, but a heterogeneous network. By

assuming different levels of abstraction, we

would recognize that different kinds of entities

mediate the way we control a car: roads and

traffic signals obviously play a part, but also the

voice of a car navigation system telling us what

to do, as well as the interiorized gaze of the

authority that keeps us from breaking traffic

rules. Modifying any of these actors might

produce a different driving experience and

different subjectivities.

The autopilot illustrates how the introduction

of a new mediator might change the way

humans perceive the environment and act. The

control of the car is delegated to software,

which dissociates the body of the human driver

from the movement of the vehicle. The car

keeps driving on its own. However, the self-

driving systems currently available have not

yet reached full autonomy and from time to

time require the human driver to take control

of the vehicle. The human driver is now

supposed to assume the role of “drive

monitor”, ready to take control of the car

whenever the software is perceived to be doing

something wrong. Given this new role, it seems

that in our example both the human and the

software made a mistake. The human was

wrong since he did not correct the autopilot.

Yet, it has been demonstrated that the high

level of automation of current software lets

humans disengage from the driving task, to

such an extent that they become potentially

less attentive and therefore unable to take

control when necessary (Banks et al 2018). Until

recently, driving has been a bodily experience:

dr ivers rece ived a feedback at every

movement, and contextual cues activated

habitual responses. The autopilot does not just

change the car; it also changes the human

driver as a subject. The way we look and feel

the road is different.

2.3 Prescription

Initially employed by Madeleine Akrich

(1992), the notion of prescription conceptualizes

one of the key insights of ANT: i.e. the idea

that humans are able to inscribe programs of

action into things and act at distance. In the

process of creation of an artefact, designers

envision the way the artefact is assumed to

interact with us and inscribe such vision in it.

“Like a film script, technical objects define a

framework of action together with the actors

and the space in which they are supposed to

act” (Akrich 1992, 208).5 Software is the most

obvious example of prescription - it literally is a

collection of scripts encoded by a programmer -,

but the concept can be extended to any kind

artefact even outside the realm of ICT. Latour


gives the example of the speed bump (Latour

1999, 185), a very simple technology used to

make drivers slow down. In this case the script

is materially built into the road.

The notion of prescription has several ethical

implications. Scripts might be implemented to

produce “moral delegations” (Latour 1999, 217).

Designers can inscribe moral instructions into

technologies, in order to make users behave

accordingly. This commonly occurs through

patterns of punishment and rewards. The

acoustic signals and visual messages produced

by a car, telling the driver to fasten his seatbelt

- or, in more recent cars, to hold the steering

wheel while a self-driving system is enabled -,

are examples of moral delegation.

B u i l d i n g o n t h i s n o t i o n , t h e

postphenomenologist Peter Paul Verbeek

stresses the necessity for a proactive approach

to technology. Given “that technologies

inevitably play a mediating role in the actions

of users”, what we need to do is to moralize

our technological environment (Verbeek 2006,

377). But who is supposed to moralize our

technologies? According to Verbeek, the script

approach “reveals a specific responsibility of

the designer, who can be seen as the inscriber

of scripts” (Verbeek 2006, 362). It is thus the

duty of designers to anticipate technological

mediat ions and moral ize technology in

accordance.

Design becomes a political matter and has

fundamental moral implications. However, we

should not forget that designers do not act

autonomously . They are part of larger

networks. Designers typically operate inside

organizations and follow decisions taken

e lsewhere – e .g . the product p lanning

department. Moreover, the notion of “good” is

context -dependent and might d iverge

fundamentally depending on the aims of the

organization in which designers operate. From

the point of view of a car maker, the best

autopilot is not necessarily the safest one.

Indeed, at the current technological stage, the

safest autopilot would probably be one that

needs the human driver to never release the

steering wheel and be constantly attentive.

Howeve r , t h i s wou l d unde rm i n e t h e

convenience of the system itself, making it less

appealing to consumers, something which the

CEO of the company that manufactured the

car mentioned in the introduction clearly

recognizes: “This is crux of matter: can’t make

system too annoying or people won’t use it”.6

A few months after the accident the same

company released a software update that

strongly increased the frequency of the “hold

the steering wheel” alert signals whenever a

driver is not touching the steering wheel.7 As

far as we know, a different prescription might

have made the driver take control of the car in

time, thus avoiding the accident. Yet, blaming

the designers for letting the driver get too

complacent would be inaccurate, since it

neglects the larger network that determined


such design in the first place and erases the agency of the human behind the wheel.

2.4 Distributed agency

While the concept of mediation is used to

deconstruct the interaction between two actors,

the idea of distributed agency (Callon and

Muniesa 2005) allows us to analyse an event,

such as the inc ident descr ibed in the

introduction, as a network composed by a sum

of microtransactions.8 The concept in itself does

not define a special kind of agency: in principle,

any action can be seen as distributed across a

network.

Jane Bennet’s ana lys is o f the North

American blackout that occurred in 2003 is a

clear example of how this concept can be

applied (Bennet 2005). According to Bennet, the

blackout that affected 45 million people cannot

be reduced to one individual cause and no

singular entity can be held responsible for it.

Human omissions, the growing demand for

electricity by a collective of consumers, legal

deregulations, economic transactions, the

movement of electrons, etc. - the effects of all

these actions put together have collectively

contributed to the blackout . Once they

intersect, a broad set of small - apparently

negligible - actions performed by different kinds

of entities, might generate huge consequences.

While the analytical advantages of the notion

of distributed agency are fairly clear, the moral

implications are more complicated. The idea of

responsibility is at stake here. In the case of

the crash, claiming that it is the assemblage

composed by the driver plus the sensors plus

the software plus the engineers plus the

stat ionary f iretruck and so on , that is

responsible for the collision of the car ends up

emptying the notion of responsibility of its

meaning, transforming it into a useless

category. It seems that ultimately nobody

should be recognized as responsible: neither the

manufacturing company, nor the engineers that

developed the software, nor the owner of the

car, nor the law, nor the software itself. Bennet

goes as far as to claim that the idea of “strong

responsibility” is empirically false (Bennet 2005,

463). However, even this assumption does not

necessarily imply an incompatibility between

the concept of distributed agency and moral

analysis. “A distributive notion of agency does

interfere with the project of blaming, but it

does not thereby abandon the project of

identifying (what Arendt called) the sources of

harmful effects” (Bennet 2005, 463). Bennet is

quietly suggesting that moral analysis does not

necessarily coincide with responsibility-

assignment. An idea that – as we will see - is

also at the centre of Floridi’s work on

information ethics.


2.5 Ethical issues of ANT

As Verbeek has stressed in his work, the

idea that technological mediations shape our

perception of reality and frame our behaviours

is a major contribution of ANT to the field of

ethics (Verbeek 2005, 2006). According to

Verbeek, this insight encourages a proactive

approach to e th ics : g iven that pure ly

autonomous humans do not exist and that our

moral decisions are necessarily mediated by

technologies, the new task of ethics is that of

moralizing our material environments. But

what moral orientation should designers follow

in this quest for moralizing technologies? What

actors should they consider as the moral

patients of their work? Humans, biological

organisms, organizations? ANT cannot tell us

what is good and what is wrong, or how we are

supposed to act in specific situations. This is

because ANT does not assess which entities

should be considered as moral patients - i.e. the

objects of moral action. This is not necessarily

a limit, since ANT does not aim at producing a

normative theory of ethics. Rather, it is

supposed to provide an agnostic tool-kit for

ethical analysis, which, in principle, could be

adopted by different ethical theories. According

to ANT’s flat ontology, any theory of ethical

patientness would in fact be equally contingent.

The limit of ANT is not so much the lack of

a definition of moral patients, but rather the

lack of a clear definition of the category of

moral agents. An army of drones and a tsunami

are both actors that could kill people, but are

they both also moral agents? Following

Verbeek’s phenomenology, it could be argued

that ultimately the discriminant between these

two forms of agency is human intentionality.

From this point of view, things behave morally

whenever they are human products: drones are

moral agents, while a tsunami is not. Such a

definition, however, seems to contradict the

principle of symmetry by assuming an a priori

distinction between human and non-human

actors. Due to this contradiction, this definition

partially falls back into a weak form of

anthropocentrism and does not of fer a

persuasive explanation for the behaviour of

artefacts that do not follow the script of their

creators.

I n o r d e r t o a v o i d t h e g h o s t s o f

anthropocentrism a different definition of moral

agency is needed. The next sections explore

how IE has tackled this same issue and has

produced a coherent definition of moral agency.

3 Information Ethics

Information ethics (IE) has its origins in the

late 1990s, when Luciano Floridi proposed it as

the theoretical counterpart of computer ethics

– a subject that at that time was largely


neglected by moral philosophers (Floridi 1999).

IE has emerged within the broader field of

philosophy of information. Developed in the

context of the expansion of ICT, philosophy of

information has two fundamental a ims

according to Floridi: clarifying the nature of

what we call information, on the one hand, and

investigating the possible phi losophical

applications of frameworks and methodologies

developed in the field of computer science, on

the other (Floridi 2011a, 14).

Like ANT, IE is also informed by a non-

anthropocentric understanding of reality.

However, their ontologies are based on entirely

different assumptions. ANT is a relational

ontology. Being necessarily means “being in

relation” to something. IE, instead, is based on

an informational ontology, and claims that

information is the lowest common denominator

shared by any entity (Floridi 2010, 94). This

ontological assumption leads Floridi to define

IE as an extension of ecological ethics: “all

entities, qua informational objects, have an

intrinsic moral value” and therefore have to be

taken into account as moral patients (Floridi

2010, 89).9 Thus, in contrast to ANT, IE

provides a clear definition of the category of

“moral patients”. Adopting the concept of

infosphere - the informational adaptation of the

idea of the biosphere - Floridi goes as far as to

claim that, since every informational entity is a

moral patient, the general moral principle

according to which any action should be

oriented to is that of avoiding entropy - i.e. loss

of information.

3.1 Moral agents

Adopting the method of levels of abstraction

(LoA)10, IE draws a distinction between the

category of “moral patients” and that of “moral

agents”. As already mentioned, IE considers

every informational entity populating the

infosphere as a moral patient. However, it

specifies that not every informational entity is

necessarily also a moral agent. Moral agents

are a subclass of the larger category of moral

patients. In contrast to ANT, IE clearly defines

what kind of actions and what kind of entities

can be morally qualifiable. According to Floridi

and Sanders (2004), to qualify as a moral agent,

an entity must be:

1. Interactive: it has an input and an output,

through which i t interacts with the

environment.

2. Autonomous: it has relative control over its

internal condition, so that it can perform an

action without the direct command of other

actors.

3. Adaptable: it can learn - i.e. it can change the

internal rules that determine its actions, in

r e s p o n s e t o i n t e r a c t i o n s w i t h t h e

environment.

Before examining the implications of these


criteria, it is necessary to clarify that the

definition of autonomy presented here does not

coincide with that repudiated by ANT. Floridi’s

definition of autonomy does not imply the

possibility of pure autonomous decisions, nor

does it imply the existence of some kind of

transcendental subjectivity or free will. It is a

quasi-autonomy: a partial control that a system

can exert on its internal state. Moreover,

depending on the LoA that we choose, a (quasi-)

autonomous system can be decomposed into a

network. For example, “depending on the LoA

adopted, the autopilot can be considered as a

single actor that performs the operation of

flying an airplane or as a set of interacting

actors that execute the subtasks of that

operation” (Turilli 2011, 377).

Let us now examine the implications of such

a definition of moral agency. First of all, in IE

moral agency is not limited to individual

humans, but can be attributed to biological

organisms, to organizations, and to IT-artefacts,

provided that they follow the three criteria just

mentioned. However, actions of entities that do

not meet these criteria, such as a tsunami or a

speed bump, cannot be accounted as moral. Can

an autopilot software be considered a moral

agent from this perspective? It depends. If it is

capable of making decisions and learning from

the environment - e.g. using reinforcement

learning -, yes. However, what if the car is not

able to autonomously change its internal rules,

but collects data that is then used to train the

software through supervised-learning? Even

though the car would ultimately be able to

change its internal rules by downloading and

updating the software, and even though the

code would be written mostly by machines

rather than human programmers, in this case

the individual vehicle would not be considered

a moral agent. However, we can adopt a higher

LoA, and look at the car as part of a larger

system that includes the neural networks

adopted, as well as the engineers that tweak

the software, monitor its learning process and

release updates. At this LoA, the car could be

considered as part of a larger moral agent.

A second implication is that intentionality is

not accounted as a necessary condition of moral

agency. This leads us back to Bennet’s

comment concerning “strong responsibility”

(see 2.4). Bennet claimed that, if agency is

distributed across a network, it is not possible

to appoint individual actors as moral ly

responsible for an event. A notion of moral

agency without intentionality encounters the

same obstacle. An autopilot might be able to

take autonomous decisions, but can it be

blamed for its mistakes? Even if that was the

case, due to the lack of self-consciousness,

attributing legal personhood to IT-artefacts and

punishing them in case of wrongdoing seems

completely meaningless - if not impossible

(Koops et al. 2010). This dilemma makes us face

what Matthias has called the “responsibility

gap” (Matthias 2004), a condition of increasing


distance between human creators/users and

IT-artefacts, where nobody is liable for the

wrongdoing of a machine.

Floridi is able to solve this apparent deadlock

by drawing a distinction “between moral

responsibility, which requires intentions,

consciousness and other mental attitudes, and

moral accountability” (Floridi 2011a, 88). A

distinction, which, according to the author,

finally frees normative ethical theory from the

shadow of anthropocentrism and, most

importantly, from the “regress of looking for

the responsible individual when something evil

happens” (Floridi 2011a, 88). However, while

arguing that responsibility-oriented ethics has

been unable to acknowledge the role of artificial

agents, Floridi does not dismiss the concept of

responsibility in toto. In IE humans have

responsibilities towards the whole infosphere.

They bear “ecopoietic responsibilities” (Floridi

2011a, 91) - i.e. they are responsible for the

creation and the well-being of the environment.

Similarly to what is suggested by ANT,

humans have a peculiar position in the moral

outlook of IE: they are not the only moral

agents in this world, but they are nonetheless

special due to their ability to create artefacts.

Conclusion

Focusing on the example of a self-driving car,

th is art ic le has reviewed the poss ib le

advantages, as well as the limits, of an ANT-

informed theory of ethics, and has briefly

illustrated the alternative definition of moral

agency provided by IE. Despite their different

ontological foundations, ANT and IE have

encountered s imi lar cha l lenges in the

development of a non-anthropocentric moral

approach and have often reached comparable

conclusions in tackling some of these issues.

ANT lets us recognize artefacts as moral

agents and gives us the tools to deconstruct

any event into a network. It allows us to

investigate how artefacts concretely mediate

our ethical decis ions - e .g . in terms of

prescriptions. However, it also runs the risk of

falling into a bottomless relativism according to

which nothing/nobody can ever be blamed. IE

seems to avoid this relativist deadlock by

providing a narrower definition of moral

agency , wh ich focuses exc lus ive ly on

autonomous entities capable of learning from

the environment and changing their internal

rules. Ultimately, both ANT and IE attempt to

shift the focus of ethical analysis from moral

responsibility to moral accountability. The two

theories would argue that, while a self-driving

car cannot be responsible for an accident, it

could, nonetheless, be a source of harmful

effects : i .e . it could be seen as morally

accountable. However, both ANT and IE have


not been clear enough in distinguishing and

defining these two concepts. How does “weak”

responsibility differ from moral accountability,

and do these two concepts imply the existence

of two different kinds of moral agency - e.g.

human vs. non-human moral agency? These are

issues that need further investigation.

1 Peter Valdes-Dapena, 2018, “Tesla in Autopilot mode crashes into fire truck.” CNN, http://money.cnn.com/2018/01/23/technology/tesla-fire-truck-crash/index.html, accessed 01/09/2018.

2 In this text the terms morality and ethics are used interchangeably.3 Matthew Philipps, 2012, “Knight Shows How to Lose $440 Million in 30 Minutes”, Bloomberg.

https://www.bloomberg.com/news/articles/2012-08-02/knight-shows-how-to-lose-440-million-in-30-minutes, accessed 01/09/2018.

4 The focus on ANT and Information Ethics (IE) is motivated by their explicit reference to non-human moral agency. Due to the lack of space this paper does not consider the moral implications of other posthumanist approaches.

5 Is should be stressed that in contrast to technological determinism, scripts can be more or less flexible. Users of technologies might resist against scripts by rejecting or by hacking them (Oudshoorn et al. 2002).

6 Elon Musk, 2018, Twitter. https://twitter.com/elonmusk/status/1005879049493725186, accessed 01/09/2018.7 Fred Lambert, 2018, Electrek, https://electrek.co/2018/06/11/tesla-autopilot-update-nag-hands-wheel/, accessed 01/09/2018.8 A similar idea can be found in the work of several authors influenced by ANT. Latour (1999) uses the term “composition”,

while Bennet (2005) talks about “agency of assemblages” and “distributive agency”.9 According to Doyle (2010), it is unclear why informational entities would have an intrinsic moral value.10 For a detailed explanation of the method of levels of abstraction see Floridi (2011b).

Bibliography

Akrich, Madeleine. "The de-scription of technical objects." In Shaping Technology/Building Society: Studies in Sociotechnical Change, edited by Bijker and Law, 205-224, 1992.

Banks, Victoria, and A. Eriksson, J. O'donoghue, N. Stanton. “Is partially automated driving a bad idea? Observations from an on-road study.” Applied ergonomics 68 (2018): 138-145.

Bennet, Jane. “The Agency of Assemblages and the North American Blackout.” Public Culture 17, no. 3 (2005): 445–65Callon, Michel. "Some elements of a sociology of translation." Power, action and belief: a new sociology of knowledge? London,

Routledge, 196-223, 1986. Callon, Michel, and Fabian Muniesa. “Peripheral Vision.” Organization Studies 26, no. 8 (2005): 1229–1250. Doyle, Tony. "A Critique of Information Ethics." Knowledge, Technology & Policy 23, no. 1-2 (2010): 163-75. Floridi, Luciano. "Information ethics: On the philosophical foundation of computer ethics." Ethics and information technology 1, no.

1 (1999): 33-52.Floridi, Luciano, and J.w. Sanders. "On the Morality of Artificial Agents." Minds and Machines 14, no. 3 (2004): 349-79.Floridi, Luciano, ed. The Cambridge handbook of information and computer ethics. Cambridge University Press, 2010.Floridi, Luciano. The philosophy of information. Oxford University Press, 2011a.Floridi, Luciano. "The method of levels of abstraction." The Philosophy of Information, 2011b. Floridi, Luciano. "Distributed morality in an information society." Science and engineering ethics 19, no. 3 (2013): 727-743.Himma, Kenneth Einar. "Artificial agency, consciousness, and the criteria for moral agency: What properties must an artificial

agent have to be a moral agent?" Ethics and Information Technology 11, no. 1 (2009): 19-29.Hume, David. A Treatise of Human Nature. Oxford: Clarendon Press, 1896.Koops, Bert-Jaap, Mireille Hildebrandt, and David-Olivier Jaquet-Chiffelle. "Bridging the accountability gap: Rights for new entities

in the information society." Minn. JL Sci. & Tech.11 (2010): 497.


Tommaso　Barbetta（とまぞ・ばるべった）

［生年月］ 1990 年 5 月［出身大学または最終学歴］ヴェネツィア・カ・フォスカリ大学　アジア・北アフリカ研究家修士課程修了［専攻領域］ STS, プラットフォーム・スターディーズ［所属］東京大学大学院学際情報学府博士課程

Latour, Bruno. "Give me a laboratory and I will raise the world." Science observed (1983): 141-170.Latour, Bruno. "Where Are the Missing Masses? The Sociology of a Few Mundane Artefacts" (1992).Latour, Bruno. "On actor-network theory: A few clarifications." Soziale welt (1996): 369-381.Latour, Bruno. Pandora's hope: essays on the reality of science studies. Harvard university press, 1999. Latour, Bruno. "Morality and technology." Theory, culture & society 19, (2002): 247-260.Law, John. "Notes on the theory of the actor-network: Ordering, strategy, and heterogeneity." Systems practice 5, no. 4 (1992): 379-

393.MacKenzie, Donald. "A sociology of algorithms: High-frequency trading and the shaping of markets." Unpublished paper (2014).Matthias, Andreas. "The responsibility gap: Ascribing responsibility for the actions of learning automata." Ethics and information

technology 6, no. 3 (2004): 175-183.Oudshoorn, Nelly, Ann Rudinow Saetnan, and Merete Lie. "On gender and things: Reflections on an exhibition on gendered

artifacts." In Women's Studies International Forum, vol. 25, no. 4, pp. 471-483. Pergamon, 2002.Peterson, Martin, and Andreas Spahn. "Can technological artefacts be moral agents?" Science and Engineering Ethics 17, no. 3 (2011):

411-424. Pitt, Joseph C. "“Guns Don’t Kill, People Kill”; Values in and/or Around Technologies." In The moral status of technical artefacts,

pp. 89-101. Springer, Dordrecht, 2014.Schüll, Natasha Dow. Addiction by design: Machine gambling in Las Vegas. Princeton University Press, 2012.Turilli, Matteo. “Ethical Protocols Design”, in Machine ethics. Edited by Anderson, Michael, and Susan Leigh Anderson, 375-397,

Cambridge University Press, 2011.Verbeek, Peter-Paul. What things do: Philosophical reflections on technology, agency, and design. Penn State Press, 2005.Verbeek, Peter-Paul. "Moralizing Technology: on the morality of technical artifacts and their design." In paper for workshop,

Utrecht University (24 pp). 2006.World Health Organization. “Global status report on road safety 2015”.


　　Smart-technologies substituting us and acting for us have become increasingly ubiquitous

over the last decade. We are quickly getting used to delegating everyday tasks to a multitude of

artefacts. Artefacts that constantly mediate our experience of reality and help us in the process of

making decisions. However, from a moral point of view, we do not yet know how to consider all these

entities.

　　In the social sciences, Actor-network Theory (ANT) has provided a consistent framework for

the analysis of non-human agency. This has been theoretically possible thanks to the detachment

of the notion of agency from that of human intentionality. However, it is not clear if and how a

notion of agency detached from intentionality could also be embraced by the field of ethics. What

is the usefulness of ascribing moral agency to non-human entities? Would such a new notion of

moral agency be a characteristic of one specific category of entities, or could it be ascribed to

anything? Furthermore, how could the question of responsibility be reframed in order to fit a non-

anthropocentric ethical approach?

　　The paper focuses on the crash of a self-driving car, an example which is used to review

advantages and limits of an ethical framework informed by ANT. Moreover, the article illustrates

the alternative non-anthropocentric approach of Information Ethics (IE), highlighting the potentials

of its narrower definition of moral agency. Ultimately, the paper shows that, despite their different

ontological foundations, ANT and IE reach comparable conclusions in the moral analysis of the car

crash: both these theories leave in fact the door open for the assessment of the moral agency of a

self-driving system. This is possible due to a conceptual shift from the idea of moral responsibility to

that of moral accountability, terms which, however, still lack a fully consistent definition.

* University of Tokyo, Graduate School of Interdisciplinary Information Studies, ITASIA D2

Key Words：Actor-network theory, Information ethics, posthumanism, moral agency, responsibility, morality of technology.

Is My Car Evil? A Review of Non-Anthropocentric Theories of Moral Agency

Tommaso Barbetta*

Documents

Is My Car Evil? A Review of Non-Anthropocentric Theories of ......Is My Car Evil? A Review of Non-Anthropocentric Theories of Moral Agency 107 Introduction January 22, 2018. It is